Cristina Luca

Data integration approach for semi-structured and structured data (Linked Data)

Researchers in the semantic web community have been designing tools and architectures to integrate heterogeneous data originated from distributed sources in the last decade. Technologies, such as RDF, have enabled new data spaces and concept descriptors to define an increasing complex and heterogeneous web of data. To these growing data resource spaces it is necessary to bridge between the different knowledge domains and provide a homogeneous and user-friendly interface. This paper discusses the limitations of some of the existing tools and solutions. The authors also propose a new architecture to integrate, using a mediator and centralised global schema, semi structured heterogeneous sources with large scale of structured data sources, particularly Linked Open Data cloud, and provide a search interface for non-expert users to searching and to accessing to them.

User experience and efficiency for semantic search engine

Search tools are essential for an information system. Being syntax based, the existing search engines have a number of limitations in particular their difficulty in returning relevant results. Typically these limitations are partly mitigated through use of various customised techniques in order to provide the best possible user experience. Unlike conventional search engines, a semantic web search engine attaches meanings to key words. One of the biggest challenges for such a semantic search engine is to maintain the standard of user experience while serving its purpose of finding relevant data with meanings. A semantic search tool needs to both link meanings to search keywords and be able to efficiently retrieve information. The purpose of this work is to investigate the techniques that can be used for efficient information retrieval and effective user experience within the context of a semantic search engine. We propose two algorithms for query optimisation and ontology crawler.

New Framework for Semantic Search Engine

The semantic web is a technology to save data in a machine-readable format that facilitates machines to intelligently match that data with related data based on meanings. Whilst this approach is being adopted and implemented by some large organisations there is a need for an effective semantic search engine to maximise the full potential of that semantic web. A major difficulty is that the search experience is dependent on a number of elements including a user-friendly interface, a strong query language processor, a result optimiser, result ranking and the use of appropriate data structures to store data. Apart from the technical aspects related to implementation, a strategy to prioritise these elements is needed to optimize and enhance the search experience over the semantic web. The purpose of this work is to investigate some relevant issues on querying the semantic web in a context of semantic search engines, and propose a framework that facilitates an effective search over the semantic web.

An interlinking approach based on domain recognition for Linked Data

The problem of interlinking datasets in, and with the Linked Open Data cloud has been one of major challenges and an important research subject in the Semantic Web. The interlinking is frequently addressed on existing data. The proposed approach, however, suggests a system to verify in first place the existence of the Uniform Resource Identifier of the resource being published in the cloud in order to establish links with it. The overall aim of the research is designing a system to facilitate the follow of the best practices and recommendation in publishing data into the Linked Open Data cloud. The main contribution of the presented paper is the use of the domain to allocate variable weights in measuring the similarity of the instances, according to the significance of their properties in defining the identity of the dataset.

Free-text user queries for semantic search

The amount of linked open data is growing fast. Yet, it is difficult for the end users to access the semantic data without having the in depth knowledge of the SPARQL language. The modern semantic search engines tackled the problem with standard user interfaces. However, these interfaces are not very usable for people with no understanding of semantic technologies. It is challenging for the semantic search tools to meet the criteria of a user-friendly search interface and the biggest challenge is to translate informal free text user queries to formal queries that can be resolved by a query processor i.e. SPARQL. In the present work the authors introduce an interface based on their previously modelled framework. The interface translates end user queries to SPARQL syntax by tagging the search terms with related concepts and properties. The authors explain the implementation of the search framework followed by the evaluation of the process.

Web-service framework for environmental models

Environmental scientists always use different models to simulate the real world with their own devices. Because of the limited computing power, the simulation can consume very long time. Also, it is hard to share their models with others and they may need to rebuild the same model for similar problems. To solve these problems, we propose a web-service centric framework to expose models as services in this paper. The framework allows model execution in the cloud environment, submission of model data through the NetCDF standard data format and storage and access to the model resources through web services. The framework allows an easy way to publish models as Linux container images through an image hub. A prototype is introduced and implemented to prove the idea works.

Schema Matching for Semi-structured and Linked Data

The Linked Data paradigm is a common standard initiated to complement the general architecture of the semantic web and create a single space containing data that is machine-readable and connected to related data. Figures, however, show that one of the main sources of semi-structured data providers, Web APIs, continued to grow even after the creation of the Linked Data concept. Given that data sources with significant value are still in a semi-structured format, it is essential to bridge between the two data models, so that the full potential of the semantic web can be realised. This paper presents SimiMatch, an approach for schema matching between semi-structured and Linked Data. It contributes towards a virtual integration system that will be able to provide transparent access to heterogeneous and autonomous sources. It addresses the challenge of sustaining the continuous changes of the web of data via semantic similarity measurement.

Minimisation of parameters for optimisation of Algorithmic Trading Systems

Many existing Algorithmic Trading Systems (ATS) use a large number of parameters to improve the results (i.e. accurate market forecasting and maximisation of profit). However, the use of superfluous parameters can often mask underlying trends leading to result instability and an excessive risk. This paper proposes a method for choosing the minimum number of parameters that are necessary and sufficient for the ATS to demonstrate a more reliable outcome. An ATS system is implemented and tested with the method and is shown to be successful for a number of case data scenarios.

Using semantic similarity for schema matching of semi-structured and linked data

This paper presents SimiMatch, an element based schema matching approach for semi-structured and Linked Data. It contributes towards a virtual data integration system that is able to provide transparent access to heterogeneous and autonomous sources. SimiMatch addresses the challenge of sustaining the continuous changes of a large-scale web of data through the use of semantic similarity measurement. The output is a domain-dependent global schema that is created and updated automatically through an unsupervised process. A test of the implementation and an evaluation of SimiMatch in three domains (movie, geographical and people data) demonstrate the effectiveness and the performance of the approach.

Result Optimisation for Federated SPARQL Queries

Linked open data end-points are accessible to users and applications to exploit semantic data. However it is not very usable for non-expert users. A search tool can be evaluated on certain criteria i.e. Access to knowledge resources, usability, expressivity and scalability. For a user friendly semantic search tool, there is a number of challenges e.g. Translating user query to formal SPARQL query, scalability, guidance, optimising query results for better readability and visualisation etc. The focus of this paper is the result optimisation for the data received as a result of running queries over different SPARQL end-points. The paper highlights two aspects of the result optimisation i.e. (i) result ranking and (ii) result readability. The paper introduces algorithm for result ranking and suggests the layout schema that defines the result template for the user interface.